Demand for AI data center construction remains strong, and Daima favors Lumentum (LITE.US) and Coherent (COHR.US) at a time when stock prices are recovering

The Zhitong Finance App learned that on Friday, Morgan Stanley pointed out in a newly released research report that although artificial intelligence (AI) data center expansion still faces multiple challenges such as power supply, network connectivity, and infrastructure efficiency, the demand for computing power from hyperscale cloud computing vendors is still strong. As the scale of AI clusters continues to expand, the importance of Ethernet, optical communication and related test equipment has further increased. The bank is optimistic about the development opportunities of companies such as Arista Networks (ANET.US), Cisco (CSCO.US), Lumentum (LITE.US), Coherent (COHR.US), and Keysight Technologies (KEYS.US) in the AI infrastructure upgrade process.

The Morgan Stanley research team attended the 2026 Open Computing Project Global Summit held in San Jose, California, USA from October 12 to 15. The conference attracted about 15,000 participants, and about 3,000 potential participants were unable to enter due to venue capacity restrictions, reflecting the AI infrastructure industry chain's strong interest in next-generation computing and network technology.

However, the bank pointed out that compared with previous years, the new industry information brought by this year's OCP conference was relatively limited. The main challenges facing data center expansion, including power supply capacity, network architecture, and optical communication technology routes, have been widely discussed at previous industry conferences.

Despite this, Morgan Stanley believes that this conference once again revealed an important trend, that is, hyperscale cloud computing vendors are still actively seeking to expand the scale of computing power, and demand for electricity, GPUs, CPUs, network connections, and optical communication devices is still strong.

Ethernet accelerates its entry into the AI cluster Arista and Cisco are expected to benefit

Morgan Stanley believes that AI network infrastructure is shifting from simply pursuing higher bandwidth to simultaneously focusing on computing resource utilization, network stability, and overall operational efficiency. As the scale of AI training and inference clusters continues to expand, the network not only needs to connect more computational accelerators, but also needs to reduce system interruptions and improve the actual utilization efficiency of existing GPUs and CPUs.

In this context, Ethernet is gradually expanding from traditional data center networks to scale-up networks within AI clusters. The scale-up network is mainly used to connect multiple accelerators within the same computing system or cluster, so that AI workloads can be efficiently processed collaboratively.

Morgan Stanley pointed out that as scale-up-enabled Ethernet switches begin to be shipped and the industry continues to push for unified network standards, the application prospects of Ethernet in AI infrastructure are becoming more clear. This will expand the potential market space for Arista Networks and Cisco.

However, the bank also warned that the promotion of the Ethernet architecture does not mean that Nvidia (NVDA.US)'s competitiveness in the AI network market will inevitably decline. Nvidia also supports various network operating systems and is expanding its network technology ecosystem through collaboration with industry chain partners. Therefore, the key to future industry competition is not only what kind of technical standards are recognized by the market, but also depends on the mass production deployment of specific products and which equipment suppliers customers ultimately choose.

In addition, the growing demand for AI inference is also expanding the scope of application of network devices.

At the conference, Nvidia presented a storage network optimized for key-value caching (KV Cache) and a scale-in network architecture connecting users, models, and data. Morgan Stanley believes that these emerging applications are expected to create more opportunities for Arista and Cisco, so that their potential market is no longer limited to high-speed interconnection within AI training clusters.

Optical communication upgrades are imperative Lumentum and Coherent usher in new opportunities

In addition to Ethernet, optical communication technology is also an important focus of discussion at this OCP conference. As the scale of AI clusters expands, traditional electrical connections face increasing limitations in transmission distance, power consumption and bandwidth density, and the application of optical interconnection technology within AI systems is becoming more important.

Morgan Stanley believes that the penetration of optical communication into scale-up networks has become a long-term trend, and near-package optics (NPO) and co-packaged optics (CPO) will be important technical routes driving this change. Among them, NPO places optical devices close to computing or switching chips, while CPO further integrates the optical engine and processing chip in the same package system to shorten the transmission distance of electrical signals and improve power consumption performance.

However, at present, the industry has not reached a unified conclusion on the best application scenarios for NPO and CPO.

According to LightCounting data quoted in the research report, customers more closely linked to the Nvidia ecosystem may have adopted CPO earlier, while other large cloud computing companies are still actively exploring different technical solutions such as NPO. The agency anticipates that by 2031, CPO and NPO will become the main forms of technology in the 3.2 Tbps high-speed port market with a single channel rate of 400 Gbps.

Morgan Stanley believes this trend is beneficial to optical communications companies Lumentum and Coherent. However, the overall expansion of the optical communications market does not mean that all suppliers will receive the same proportion of revenue growth. The laser configurations, number of optics, and supplier choices used in different architectures will still influence the extent of ultimate benefits.

In the context of the recent sell-off of related stocks, the bank is particularly concerned about investment opportunities in Lumentum and Coherent, and believes that subsequent commercialization of NPO technology is expected to become a new catalytic factor for the two companies. However, the bank stressed that this year's OCP conference did not substantially change its previous investment judgment on the two companies.

Meta Announces CPO Reliability Progress, Laser Supply Is Still the Key

In terms of commercialization of CPO technology, the latest test results disclosed by Meta Platforms (META.US) have received attention. According to Morgan Stanley, Meta said its more mature laboratory CPO system has improved interruptions by about 6 times compared to traditional pluggable optical modules. At the same time, in tests with a cumulative total of 1.35 million hours of system operation, there were no faults in the relevant systems that could not be repaired through maintenance.

The bank believes that this result enhances the viability of future large-scale deployment of CPO technology, but different vendors and subsequent product platforms still need to complete reliability verification separately, and existing test results cannot be directly promoted to the entire industry.

At the same time, traditional pluggable optical modules will not be quickly withdrawn from the market.

Nvidia mentioned during the Q&A session of the conference that USFP products are expected to enter production in 2027 and listed Cisco as one of the relevant development partners.

Morgan Stanley notes that optical communication solutions based on electrically absorption modulated lasers (EML) still face supply chain challenges. At present, the supply of 200G EML is tight. As the industry further develops towards 400G EML, the difficulty of procuring related products may increase.

In terms of external lasers, Lumentum said its ELSFP module is already in production. However, whether to place the laser on the device panel or integrate it into the system is still an important issue of debate in the industry. The former is beneficial for on-site maintenance and device replacement, while the latter may have advantages in terms of power consumption and assembly complexity.

The bank believes that investors need to pay attention to the laser solutions eventually adopted by different platforms and the value of optical devices included in each architecture, rather than simply assuming that the CPO market growth will benefit all laser suppliers simultaneously.

Demand for connecting AI clusters across data centers is growing, and Ciena and Keysight Technologies are also receiving attention

As AI computing tasks become more complex, some cloud computing vendors are exploring cross-data center collaborative computing, or scale-across network architectures. This model requires connecting computing resources in different geographical locations, and places higher demands on network latency, congestion management, and fault recovery capabilities.

Morgan Stanley believes that interconnection across data centers will increase the demand for long-distance optical transmission equipment, which is particularly beneficial to Ciena (CIEN.US). At the same time, different vendors are exploring flatter network structures to reduce switching levels, reduce power consumption, and improve network efficiency.

For example, the multi-plane network architecture presented by Meta reduces the switching layer from three to two layers, while hardware fault detection and traffic rebalancing techniques introduced by Nvidia can reduce related processing times to less than 3 ms.

Morgan Stanley pointed out that these technological advances mean that future growth in the network equipment market cannot be judged only by bandwidth requirements; it is also necessary to consider the number of network devices required for each accelerator and changes in the overall system architecture.

Furthermore, as AI data centers are simultaneously testing various network and optical communication solutions, the demand for testing and verification of related equipment will also increase. The bank believes that Keysight Technologies is expected to benefit from the increasing diversification of optical communication architectures and the expansion of high-speed optical device testing scale.

There is still room for growth in the AI network market, but vendor competition will be more complicated

Morgan Stanley believes that the 2026 OCP conference did not bring about technological breakthroughs sufficient to completely change market expectations, but it further confirms that demand for AI infrastructure is still strong. As hyperscale cloud computing vendors continue to expand their computing power deployments, network connectivity, power efficiency, and optical communication technology will become important factors in determining the speed of AI data center expansion.

In terms of specific investment direction, the bank believes that the short-term expansion of the Ethernet scale-up market is most positive for Arista Networks and Cisco, while the continued development of NPO and CPO technologies provides new growth opportunities for Lumentum, Coherent, and Keysight Technologies.

However, the bank warned that the industry is promoting more open network standards and interoperable hardware platforms, and customers also want to reduce the number of devices and system power consumption. As a result, the expansion of the AI network market does not necessarily translate directly into the same proportion of revenue growth for all equipment vendors.

In the future, whether suppliers can obtain orders in specific customer deployments, maintain a high share of product value in new architectures, and continuously improve equipment energy efficiency and operational stability will be the key to determining their long-term competitiveness.